Miko
An open-source AI desktop companion for Windows that uses Ollama, custom VRM avatars, and either offline voice cloning or Fish Audio cloud speech.
Miko is an AI companion for people who want to talk with an animated character on their Windows desktop. It pairs a local language model with a customizable 3D avatar and spoken replies.
- Desktop avatar: Use your own VRM character with lip-sync, blinking and gestures in a transparent overlay that lets clicks reach the applications underneath.
- Voice and text conversations: Talk through push-to-talk or continuous microphone listening, or type messages into a console; spoken replies can begin before the model finishes its response.
- Local or cloud speech: GPT-SoVITS provides offline voice cloning on your hardware, while Fish Audio generates expressive multilingual speech in the cloud without using local GPU memory for speech synthesis.
- Conversation memory: Miko recalls relevant past exchanges, gives preferences and names greater weight, and reduces the prominence of older memories and repeated statements.
- Device control: The companion supports smart-home IoT control and OASIS physical turntable hardware integration.
Miko runs on 64-bit Windows 10 and 11 and is open source under GPL-3.0. Its language models run locally through Ollama, with Llama 3 and dolphin-llama3 among the named choices. Choosing Fish Audio requires an internet connection and an API key, and sends speech generation to that service; the language model still runs through Ollama.